What do Oxford want?

Visualising rabbit hearts

Integrative Biology

e-Science project investigating the causes of heart attacks how cancers develop

Together these account for 60% of all UK deaths

The IB project brings together numerical modellers, experimentalists, computer scientists, and even mathematicians

The Oxford Conference

Everybody meets up each September In 2005 Ken and James (Handley) presented the

multivariate visualisation work In 2006 the other half of the computer scientists

contingent (Lakshmi Sastry from CCLRC) presented the Grid computing environment they’ve built (using gViz amongst others)

At the questions stage Peter Kohl from Oxford asked

“Aren’t you supposed to be providing cool visualisations for us?”

Who is Peter Kohl?

Marital status: Married, three children

Nationality: German, resident in UK since August 1992

Languages: Fluent in German, English and Russian

Hobbies: Preparing (and having!) a good dinner and landscape photography

Who is Peter Kohl?

Reader in Cardiac Physiology, Department of Physiology, Anatomy and Genetics, Oxford University

Head of the Mechano-Electric Feedback lab

“The purpose of our group is to incorporate kinetic properties of molecular and cellular mechanism of cardiac activity into cellular and multicellular mathematical models of the heart.”

What data does he have?

Five valiant bunnies put their hearts into this research

They were scanned using an 11.7Tesla magnet to give about 1500 slides, each 25m with resolution 1024x1024 points.

The hearts were then sliced to a thickness of 10m, stained, and photographed using a scanning microscope (1800 slices, resolution of up to about 6000x4000)

MRI data

MRI data

What we’ve done already1. MRI data VTK

Loads and isosurfaces Save pre-computed isosurfaces for later rendering

Powerwall vjVTK

Performance not good enough – especially on the headnode

Homegrown OpenGL version in VRJuggler Smoothing and decimation of the original VTK

surfaces

Histology data

What we’ve done already2. Histology data Slide viewer

Desktop OpenGL and VRJuggler versions Slides much smaller than the pathology ones so not using

the multi-tiled textures Multiple slices viewing

Individual slides need much preprocessing Adding transprency Removing reference marker Still need to clean up the background

Can manipulate between 16 and 32 slices in 3-d Memory is the killer

What I’m working on a the minute Automatic Level of Detail operations

Using GLOD library Precomputing hierarchy of levels

Very computationally expensive on large datasets 6Mb = 1 min 24Mb = 30 mins 40Mb = 3hrs

…hence need to break the geometries down into smaller sections This will also help at display time

What do Oxford want?

To combine the histology with the MRI Complete data registration

between slices To the MRI set

Need to decide on what viz Adding one slice every 20-100? Colour isosurface by histology value? Showing histology in a separate

window matching cut plane

Cool things that are probably too hard Volume rendering

Memory requirements are too large May work for one off pictures to be preprocessed Knowledge of suitable transfer functions needed

3-d texture environments Again, enormous memory requirements No segmentation done (that we know of)

Cool things that should be do-able Heart Atlas

Multiple hearts loaded across the display Comparisons between structure recognisable

More ideas needed…

We want these techniques to scale to hearts that are up to 20 times larger!