Motion-Moderated Transfer Function for

Volume Rendering 4D CMR Data

(alt. title: what did rainbow colourmaps ever do to you?)

Simon Walton1 and Min Chen1 and Cameron Holloway2

1 Oxford University 2 St. Vincent's Hospital, Sydney

Overview

• Brief introduction to CMR

• Motivation for motion-moderated TFs

• What are motion-moderated TFs

• How we implemented them

• More Results

Cardiovascular MRI (CMR)

Cardiovascular MRI (CMR)• Essentially ECG-timed MRI

https://www.med-ed.virginia.edu/courses/rad/cardiacmr/Techniques/

Gating.html

beepbeepbeep

CINE Imaging

LV

RV

apex

mid

base

Slices often targeted

Colourmapping

low density high density

Problems with Colourmaps• The colour bands are mapped

to raw intensity

• Intensity values represent density and not 'tissue type'

• Myocardium density changes through systole and diastole

• Therefore, intensity-mapped colour alone cannot visually track tissue

t distant distractor

t+1

nearby distractor

target region

distant distractor

nearby distractor

target region

We wanted a solution which...

• ...could help alleviate density incoherence

• ...didn't require a new capture technique

• ...maintained the primary method of viewing

• ...could benefit 4D volume rendering generally

New Transfer Function?• Transfer function f(x) = t

• A couple of examples of transfer function input types in the literature:

• Size, visibility (Correa et al. 2008; 2009)

• Histogram (Caban et. al. 2008)

• Shape (Sato et al. 2000)

• How about motion?

Motion-Moderated Transfer Functions

A motion-moderated TF

• We can use motion information as input

• Can bring the attention to areas under motion

• Can aid with occlusion caused by static objects in the foreground

• Can aid identification of high-velocity areas

t1

t10

GradientIntensity Motion Mag

How to obtain motion metadata

• Phase-contrast velocity mapping

• Tagged CMR techniques (see right)

• However:

• We cannot rely on this

• We wish for generality

Optical Flow estimation• For generality we use

optical flow estimation

• Output is a 3D grid of vectors for each time step

Workflow

1 bu

ild compute motionintensity

t

intensityt-1

motion field

Intensity Field: original volume dataset Motion Field: generated by optical flow

Raycasting to 2D buffers

Intensity Field and

Motion Fieldcolour buffer

c: colourmap(intensity)α: motion mag

Motion Field

motion buffer

( )

Workflow

1 bu

ild compute motionintensity

t

intensityt-1

motion field

2 ra

ycas

t

accumulate colour

accumulate motion

Problem: noisy motion• The motion vectors from optical flow techniques

are usually noisy

• E.G. if we follow a region from left to right:

t0 t1 t2 t3 t4

• This can produce artefacts in the imagery

• Motion from frames t-1, t-2, etc can be used to smooth out the motion in t

++

t

t� 1

t� 2

t� 3+

Solution for noisy motion

Workflow

1 bu

ild compute motionintensity

t

intensityt-1

motion field

2 ra

ycas

t

accumulate colour

accumulate motion

3 co

mpo

se

temporal coherence

framebuffer

Results

intensity intensity gradient motion mag x

intensity

motion

motion x

intensity

motion x

intensity

motion2 x

intensity

Alpha from intensity values

Alpha from motion magnitude

Conclusion• We have introduced a transfer function utilising

motion information

• Such a technique can be generally applied

• Relies on an optical flow technique

• Clinical data often of reduced quality

• Research patients can be scanned for longer

Thanksfrom the heart (groan)