volume flow determination in the cranial vessel tree based on quantitative magnetic resonance data

Volume Flow Determination by QMRA / 2008/02/15 of 64

Volume Flow Determination in the Cranial Vessel TreeBased on Quantitative Magnetic Resonance Data

Supervisor (TUM):

Supervisors (BrainLAB):

Andreas Keil

Thomas Seiler,

Fritz Vollmer

Advisor: Prof. Dr. Navab

by Jürgen Sotke


• Goal

• Quantitative Magnetic Resonance Angiography (QMRA)

• State of the Art

• New Approach

• Results

Agenda


Goal


Quantitative information about volume flow rates(either abstract or graphicaly)

Purposes:

• Diagnosis(stenosis, ischemia)

• Review ofoperation results

Goal


QMRA


So far there exists only one MR technique which allows to directly measure flow velocities:

In a phase contrast image, the grey level is linearly dependent to

the velocity of the blood.

Phase Contrast Image:bright = high velocities in the direction of the scandark = high velociteis in the opposite direction

phase contrast MR

QMRA


Two undesired effects:

1. Limited velocity range

2. Works only for blood flow in one given direction

QMRA

Lotz J., Meir C., Leppert A. et al.: “Cardiovascular Flow Meaurement with Phase-Contrast MR Imaging: Basic Facts and Implementation”, RSNA, 2002


State of the Artor...

State of the Art

Volume Flow Determination by QMRA / 2008/02/15 of 64State of the Art

http://www.youtube.com/watch?v=a7rBJWhCkF8&feature=related

http://www.vassolinc.com/QuickTourNOVA.cfm

Visite

or

for a video about the current use of QMRA.


Pre-planed slices

- inefficient workflow- requires registration- only flow information for a few samples

State of the Art

http://www.vassolinc.com


Agenda

• Goal

• QMRA


• The New Approach

• Results


Combining an abstract model of the vessel treewith flow information.

New Approach


Agenda• Goal

• QMRA



• Results

1. Data Acquisition

2. Segmentation

3. Creation of an Abstract Model of the

Vessel Tree

4. Adding Flow Information to the Abstract Tree

5. Improving Flow Information by the Use of

Topological Information


Data Acquisition

New Approach/Data Acquisition


Only one session

with PCA scans in at least three orientationsover the whole volume.



Because of pulsatile fluctuations, some kind ofaveraging over the heart beat is necessary:

Each plane consist of a set of PCA slices depicting flow during different intervals of the (ECG-triggered) heart beat cycle.



Agenda• Goal

• QMRA



• Results

1. Data Acquisition

2. Segmentation


Vessel Tree




New Approach/Segmentation


Segmentation directly from the phase contrast datarequires combining the three orthogonal scans due to thedirectional sensitivity of phase contrast MR.


Phase contrast images only depict vessels which run roughly parallel to the scan direction


1. Combining PC-images

2. Region Growing

3. Closing

Three major segmentation steps


Eiho, Sekiguchi, S.H., Sugimoto, N. et al.: “Branch-Based Region Growing Method For Blood Vessel Segmentation”, Systems and Computers in Japan, 2005


Segmentation Result



Agenda• Goal

• QMRA



• Results

1. Data Acquisition

2. Segmentation


Vessel Tree





Creation of an Abstract Model of the Vessel Tree

New Approach/Abstract Tree Model

Segmentation result Topological Model


3.1) Topological Structure of the Vessel Tree

{

{Skeleton

{

Centerline

New Approach/Abstract Tree Model


3.1) Topological Structure of the Vessel Tree

=> Centerline-Extraction

Two common techniques:

• Distance based approaches

• Thinning

New Approach/Abstract Tree Model/Topological Structure


Distance-Transform-Map2D-object

minimal distance of the pixel to the object‘s bounds

0

1

2

New Approach/Abstract Tree Model/Topological Structure/Distance Maps

Distance Based Centerline Extraction

In the case of symmetrical 2D-objects the maxima of the DTM already pose the centerline pixels.


Distance Based Centerline Extraction


Not radially symmetrical objects possess multiple local maxima in their distance maps, which cannot be connected in a well defined way.

In 3D only radially symmetrical objects pose such distinct maxima of the distance map.

Volume Flow Determination by QMRA / 2008/02/15 of 64New Approach/Abstract Tree Model/Topological Structure/Distance Maps

Multiple Maxima in the DTM




…can be avoided by filtering the DTM


…can be avoided by filtering the DTM, but this causes a loss of connectivity in thin vessel segments.




Thinning

… works by removing the voxels at the object bounds…

… layer…

… by layer…

…until the remaining object poses only a thickness of one voxel.


Thinning Result

New Approach/Abstract Tree Model/Topological Structure/Thinning

Lamy, J.: “Integrating digital topology in image-processing libraries”, Elsevier Ireland Ltd, 2005


Centerline Extraction

Distance based approach Thinning

+ correctness

- bad connectivity

+ high connectivity

- faulty

New Approach/Abstract Tree Model/Topological Structure

=> combined approachusing centerline voxels from thinning to connect local maxima from distance transform


Centerline Extraction: combined approach

New Approach/Abstract Tree Model/Topological Structure/Combined Approach


3.2) Assignment of Volumetric Information to the Abstract Model

α α12

New Approach/Abstract Tree Model/Assignment of Volumetric Information

A voxel in the vicinity of a centerline segment is added to the assigned volume, if the two intersection angles in the image are smaller then 90°.


Agenda • Goal

• A little bit of MR-Physics



• Results

1. Data Acquisition

2. Segmentation


Vessel Tree




Volume Flow Determination by QMRA / 2008/02/15 of 64New Approach/Abstract Tree Model/Adding Flow Information

Since the three phase contrast scans are orthogonal, they can be considered as the three components of a velocity vector.


Flow Velocities in all Vessel Segments

New Approach/Abstract Tree Model/Adding Flow Information

Total flow velocities in the vessel tree.


Intersection area


The knowledge of length and volume of all segments of the abstract vessel tree allows to compute the average intersection angle in all of these segments.


Flow Rates in all Vessel Segments


With knowledge of the intersection areas, flow rates can be computed for all segments of the abstract tree.


Agenda• Goal

• A little bit of MR-Physics



• Results

1. Data Acquisition

2. Segmentation


Vessel Tree





• or generating an intelligently weighted combination of the data from different slices

Angular information allows

• selecting the most suitable PC-slice

in order to improve flow information.

Angle between sagittal plane and vessel segment 3 Angle between

coronal plane and vessel segment 3

New Approach/Improving Flow Information


232?

448

51

?

42

?

?

5048

232216

448

18151

72

42

30

98

5048

?

-

-

+

109

-

-

+

-

+

+

-

-

-

-

+-

109

Substitution of Unreliable Data in the Abstract Vessel Tree

New Approach/Improving Flow Information


Summary

QMRA poses technical limitations which so far

compelled an inefficient workflowrequiring patient or image registrationand supplying only flow information for a few selected slices.

The new approach might allow to

acquire the necessary data in a one-step workflowwithout the need for patient or image registrationthat supplies flow information for all parts of the vessel tree with an accuracy (nearly) equal to that of pre-planed slices.


Results


Segmentation from PC-Data

yespossible?

advisable? depends…

mutual improvement:future work?

• QMRA-Software would allow to detect and correct segmentation faults

• Improved segmentation would lead to improved abstract model.

Results


Visualizing Flow

yespossible?

visualizing in the abstract treefuture work?

Results


Substituting “unreliable” data in the vessel tree

not provedpossible?

would require better datafuture work?

Results


Finding a corrective factor/function

not provedpossible?

would require more datafuture work?

Results


Appendix


A little bit of MR-Physics


MRI is all about...

the angular moment (spin) of protons.


These protons prefer to align with the external magnetic field of the scanner

but can be “persuaded” (excited) to “anti-align” for a short moment.


When the protons “fall back” to the parallel state,after they were excited, they emit radio waves…


The strength of the external field has influence on how easily the protons can be excited.


The Physics of Phase Contrast


Image Plane

phaseshift

movingproton

fieldgradient

slowerproton


Two undesired effects:

1. Only 180° to encode all velocities

2. Works only for blood flow in the direction of the gradient

limited velocity range



How to get rid of them?

Averaging over close neighbors / by clusters?

=> many undesired effects

Better: preventing them by filtering


Preventing multiple maxima by filtering

Distance maps

Distance mapsafter applyinga Gaussian filter


Thinning

Thinning must take the topology of the object before removing voxels.Only voxels which are not important to preserve the objects topology are allowed to be deleted.

volume flow determination in the cranial vessel tree based on quantitative magnetic resonance data

Documents